Project conducted at BARAM (Acdaemic Robotics Club at Kwangwoon University)
QP-based MPC for Differential-Drive Mobile RobotPersonal project during Sep. - Nov. 2021 |
Experimental Results
| Real-world robot in a small maze | Visualized path planner and local MPC controller |
Real-world robot experiments to validate the QP-based MPC controller for a differential-drive mobile robot. Navigation performance of the robot is evaluated at small maze environment using sensory inputs (e.g. LiDAR and encoder).
Description
This project aims to adopt quadratic programming (QP) based local controller for the differential-drive mobile robot navigation. Commonly, the navigation system of mobile robots consists of two core components: global planner and local controller. Conventional local controller often shows sub-optimal behaviors under dynamic environments (e.g. dynamic obstacles). Optimal controllers like model predictive controller (MPC) are popular at different domains (e.g. manipulators or humanoids), while it's performance is not evaluated well at the mobile robot domain. Through this project, the effectiveness of optimal controller using QP is evaluated for real-world environments.
Conclusion
Combination of QP-based MPC controller and global planner (i.e a* planner) can navigate the differential-drive mobile robot effectively. However, heavy hand-engineered parameter tuning for MPC controller makes this controller unrealiable to various real-world scenrios. Furthermore, increased computing budget by optimal controller cause the overall operation time too slow.
Projects conducted at DYROS (Dynamic Robotic Systems Lab at Seoul National University)
Efficient map construction and navigation using multiple sensors on a single robotPersonal project during Jan. - Jun. 2022 |
Experimental Results
| When only motor encoders are used (SLAM X) | When motor encoders and LiDAR sensors are used (SLAM O) |
Real-world robot experiments to evaluate the effect of LiDAR SLAM on mobile robot localization. The occupancy grid map was pre-generated by multiple LiDAR sensors.
| Building an occupancy map using LiDAR sensor |
An occupancy grid map was generated using SICK LiDAR sensor.
Description
This project aims to justify the effect of multiple LiDAR SLAM for mobile robot localization. Simultaneous localization and mapping (SLAM) can generate precise maps using sensors (camera or LiDAR) for the navigation. However, it is still open challenge that creating high-precise map while reducing computational cost. Furthermore, the mobile robot often uses a single sensor to generate an occupancy map. Through this project, the relation between the number of sensors and the performance of SLAM is studied for mobile robot navigation.
Conclusion
Using multiple LiDAR sensors don't significantly increase the speed of map generation in general. However, under complex environments with multiple obstacles, using multiple sensors can enhance the efficiency of SLAM, and not surprisingly, the location of the sensors affects to the performance. It is important to place the multiple sensors where each sensor does not overlap as much as possible. Also, based on experiemntal results, relying on kinematic information using motor encoders shows strong performance on short-term navigation, while using SLAM shows relatively better accuracy on long-term navigation due to accumulated error of encoders.
Other awesome experiences!
| Door-opening mobile manipulation experiment | Localization experiment with motion capture sensors |